CCRL 40/15, 40/2 and FRC lists updated (26th June 2021)

Discussion of computer chess matches and engine tournaments.

Moderator: Ras

connor_mcmonigle
Posts: 544
Joined: Sun Sep 06, 2020 4:40 am
Full name: Connor McMonigle

Re: CCRL 40/15, 40/2 and FRC lists updated (26th June 2021)

Post by connor_mcmonigle »

mehmet123 wrote: Thu Jul 01, 2021 5:57 pm
lkaufman wrote: Thu Jul 01, 2021 4:49 pm
mehmet123 wrote: Sun Jun 27, 2021 7:36 pm
mehmet123 wrote: Sun Jun 27, 2021 9:32 am Great success for SlowChess Blitz.

If we don't consider NN chess engines and controversial chess engine Fat Fritz , SlowChess Blitz is now in third place.
Actually ı mean gpu based engines like Lc0 and Stoofvlees not chess engines using nnue network. Houdini is the most powerful alpha-beta chess engine that don' t use nnue network.
Stockfish 11 (or a later dev version or derivative) would be the strongest alpha-beta chess engine not using NNUE, not Houdini, which is a derivative of an earlier Stockfish (and Komodo).
I mean chess engines which no versions uses NNUE network.
Latest version of Stockfish that doesn't use NNUE networks is Stockfish 310720. I also did some tests with this engine.
http://talkchess.com/forum3/viewtopic.p ... &start=190

Yes we know that Houdini uses some Stockfish code but I don't think these Stockfish codes give Houdini as much elo points as exaggerated.
Why would anyone care that no version in an engine's history used a neural network? That seems pretty silly. Regardless, most of Houdini's strength does originate from Stockfish in that Houdini is almost entirely comprised of Stockfish code besides for some modifications to the evaluation function. The search code is effectively a verbatim copy with some changes to the probcut implementation... have you read the Houdini source code?
mehmet123
Posts: 697
Joined: Sun Jan 26, 2020 10:38 pm
Location: Turkey
Full name: Mehmet Karaman

Re: CCRL 40/15, 40/2 and FRC lists updated (26th June 2021)

Post by mehmet123 »

As far as I know, Stockfish codes were used only Houdini 6. The elo difference between Houdini 6 and Houdini 5 at 1cpu is 35 -40 elo according to CEGT and CCRL. I guess that all of this elo gain does not come from Stockfish codes.
https://PrivateLadyEscorts.com - Live Local Dating - No Verify - Anonymous Casual Dating - Chat Local Singles
connor_mcmonigle
Posts: 544
Joined: Sun Sep 06, 2020 4:40 am
Full name: Connor McMonigle

Re: CCRL 40/15, 40/2 and FRC lists updated (26th June 2021)

Post by connor_mcmonigle »

mehmet123 wrote: Thu Jul 01, 2021 6:32 pm As far as I know, Stockfish codes were used only Houdini 6. The elo difference between Houdini 6 and Houdini 5 at 1cpu is 35 -40 elo according to CEGT and CCRL. I guess that all of this elo gain does not come from Stockfish codes.
I believe it is known that Houdini 5 was also a Stockfish clone (prior versions were Robbolito derived iirc)
Terje
Posts: 347
Joined: Tue Nov 19, 2019 4:34 am
Location: https://github.com/TerjeKir/weiss
Full name: Terje Kirstihagen

Re: CCRL 40/15, 40/2 and FRC lists updated (26th June 2021)

Post by Terje »

mehmet123 wrote: Thu Jul 01, 2021 5:57 pm [...] I don't think these Stockfish codes give Houdini as much elo points as exaggerated.
You're wrong. Houdini 5 and 6 ARE stockfish, with some changes. Most of the elo is Stockfish.

And just to preempt the "but it was stronger than SF" - H5 and H6 were based on development versions of SF (between releases), some 25ish elo stronger than the at the time most recently released version of SF. There may have been a few elo gained from the changes Houdart made, but the idea that what he took from SF wasn't enormous is ridiculous.

With that said, there are some cool things in Houdini, tactical mode among others. Had he been honest about the origins and adhered to the licence (made Houdini open source), Houdini would have been a cool SF fork much like Crystal is today.